the fact that people fail the abx isn't proof that it's impossible to pass it. that much is sure, anybody who bothered to try and understand abx will acknowledge that.
but we also know a lot about statistics, and we can reliably use stats for what they are. when 50000people fail a test, and 1 comes telling he passed, you know it's possible, but you also have legitimate right to doubt the procedure and ask to know more.
because overly unfavorable odds start to make for something close to a rule.
from what I've tried and seen on the web, a mp3@320 is strictly identical to a 16/44 from 0 to mostly about -60db. (a few years back I uses to invert one track in audacity and then mix them and look at the spectra stuff. but I never really knew if it was a valid way to do it?).
anyway everybody can take a song in foobar, play it at his usual loudness with foobar volume maxed out, then lower the volume in foobar by 60db. it's no extraordinarily involving experience and it tells a lot about what differences we're actually talking about.
and to add to that, there is the masking effect, as the music playing will cleverly cover those -60db and below differences almost at all time. that's most of the work done with the mp3, to know where it can cut without us noticing because something will be masking the sound anyway at that moment.
so my opinion, from doing a good deal of mp3@320 vs lossless ABX myself, is that it is possible to succeed one, as long as you know what you're looking for, and only replay the same exact passage on the very song that has something audible.
if a guy can do it with many kinds of songs on almost any passage with relative accuracy, then I'm very very confident that he converted the file badly. it may have some clipping because the mp3 is too close to 0db. or the guy is a noob and just used 2 different masters instead of converting the file himself(lol the difference is so obvious, mp3 sucks!!!! QED \o/). or he encoded with replay gain ^_^...
or simply that his sound system sucks at converting mp3 back to pcm( I experienced that myself with the fiio X3 on the first firmware). but I wouldn't believe the mp3 tech itself could be the reason for obvious audible differences frequent enough to be noticeable without a test. so for me, anybody telling something like "I don't need an ABX to tell that lossless sounds better" is wrong!(pick a reason: liar, ignorance, placebo) that much I don't need to leave to statistics.
obviously me saying this for "loud" -60db differences, can tell you a lot about what I think of people pretending that they hear a "better" sound from 24/96 compared to 16/44... again it can happen, no doubt about it, for pretty much the same reasons as above. the encoding messed up, using 2 different masters, and probably the most obvious, any sound system that actually struggles with one of the format, like ... would distort in the audible range because of too much ultrasounds or whatever.
and so just like with cables, my opinion is that the bad users and bad device choices make up for most of the "I can hear all the betterness, and so could my wife in the kitchen". the rest being placebo. and I'm still looking for the guys the ESS people talked about who can perceive -100db changes in sound. I'm ready to believe they exist, but where are they? if I was one of those guys I would be on TV and make showy demonstrations for the highres advertising. big bucks, big celebrity, how cool would that be!!!!! sadly I'm not one of the chosen few hifi people.